Unravelling the Complexity: Understanding the Challenges of Reinforcement Learning

Brahmaleen Kaur Sidhu

Abstract: After an extensive research and exploration of supervised, unsupervised and semi-supervised machine learning algorithms, researchers across the numerous application domains of machine learning are now looking to implement reinforcement learning techniques as they promise a realization of more human-like intelligence in machines. This paper presents a comprehensive body of knowledge about the complexities and challenges that researchers might face while developing reinforcement learning models as solutions for real-life problems. Also, some recommendations have been made in order to assist effective implementation of reinforcement learning algorithms.

Keywords: computational complexity, environment specification, exploration-exploitation, reinforcement learning, safeRL, sample efficiency